ANFIS-based PCA to predict the longitudinal dispersion coefficient in rivers

Abstract

Study on the river water quality is the main part of environmental engineering. Longitudinal dispersion coefficient (DL) is one of the main important parameters in the river water quality studies. The DL is proportional to the flow velocity, channel width, depth of flow, and shear velocity. In this study the most effective parameters on DL were derived using principal component analysis (PCA). The results of PCA indicate that the width of the river, flow depth, and flow velocity are the most important parameters on the DL. In this paper with aim of the PCA results, the adaptive neuro fuzzy inference system (ANFIS) model was developed to predict the DL. The performance of developed ANFIS model based on PCA with error indices (R2 = 0.95 and RSME = 0.055) is suitable to predict the DL. During the development of ANFIS model, five neurons to the width channel, four neurons to the flow depth, four neurons to the velocity and two neurons to the shear velocity were assigned. The hyperbolic tangent sigmoid (tansig) function was considered as membership function for the neurons.